Data-driven virtual reference feedback tuning and reinforcement Q-learning for model-free position control of an aerodynamic system

Abstract

This paper compares a linear Virtual Reference Feedback Tuning model-free technique applied to feedback controller tuning based on input-output data with two Reinforcement Q-learning model-free nonlinear state feedback controllers that are tuned using input-state experimental data (ED) in terms of two separate learning techniques. The tuning of the state feedback controllers is done in a model reference setting that aims at linearizing the control system (CS) in a wide operating range. The two learning techniques are validated on a position control case study for an open-loop stable aerodynamic system. The performance comparison of our tuning techniques is discussed in terms of their structural complexity, CS performance, and amount of ED needed for learning.